Templatize the implementation

Also introduces an optimziation that uses the stack, if the buffer size
required, is lower than a pre-determined threshold

benchmarks/file-reader.cpp

@@ -0,0 +1,27 @@
+#include <vector>
+#include <string>
+#include <iostream>
+#include <numeric>
+#include <iterator>
+#include <chrono>
+#include "iosifovitch.h"
+
+int main() {
+	std::ios::sync_with_stdio(false);
+	std::vector<std::string> words;
+	words.reserve(1000);
+	std::copy(std::istream_iterator<std::string>(std::cin), {}, std::back_inserter(words));
+	std::cout << "words read: " << words.size() << '\n';
+	auto begin = std::chrono::steady_clock::now();
+	for (auto const& w : words)
+	for (auto const& y : words)
+	{
+		auto distance = levenshtein_distance(w, y);
+		if (distance > (std::numeric_limits<unsigned int>::max() - 2))
+		{
+			std::cerr << "failure : " << w << ", " << y << "/" << distance << '\n';
+		}
+	}
+	std::cout << "total time taken: " << std::chrono::duration_cast<std::chrono::milliseconds>(std::chrono::steady_clock::now() - begin).count() << std::endl;
+	return EXIT_SUCCESS;
+}

src/iosifovitch.cpp

@@ -1,40 +1,5 @@
 #include "iosifovitch.h"
-#include <numeric>
 
-auto levenshtein_distance(std::string_view const& a, std::string_view const& b) -> unsigned int {
-	if (a.size() == 0 || b.size() == 0) return a.size() + b.size();
-
-	if (a.size() > b.size()) return levenshtein_distance(b, a);
-
-	auto i = 0u;
-	while (i < a.size() && a[i] == b[i]) ++i;
-
-	if (i != 0) return levenshtein_distance(a.substr(i), b.substr(i));
-
-	i = 0;
-	while (i && a[a.size() - i] == b[b.size() - i]) ++i;
-	if (i != 0) return levenshtein_distance(a.substr(0, a.size() - i), b.substr(0, b.size() - i));
-
-	auto const buffer_length = a.size() + 1;
-	auto buffer = new unsigned int[buffer_length];
-
-	std::iota(buffer, buffer + buffer_length, 0);
-
-	for (auto i = 0u; i < b.size(); ++i)
-	{
-		buffer[0] = i;
-		auto temp = i;
-		for (auto j = 0u; j < a.size(); ++j)
-		{
-			temp = std::min(
-				temp + (a[j] == b[i] ? 0u : 1u),
-				std::min(buffer[j + 1], buffer[j]) + 1u
-			);
-			std::swap(buffer[j + 1], temp);
-		}
-	}
-
-	auto cost = buffer[buffer_length - 1];
-	delete [] buffer;
-	return cost;
+auto levenshtein_distance(char * const a, char * const b) -> unsigned int {
+	return levenshtein_distance<std::string_view>(a, b);
 }

src/iosifovitch.h

@@ -1,6 +1,7 @@
 #pragma once
 
-#include <string>
+#include <numeric>
+#include <string_view>
 
 /* Calculate the levenshtein distance between two strings.
  *
@@ -8,8 +9,45 @@
  * best case, where n is the length of the shortest string.
  *
  */
-auto levenshtein_distance(
-	std::string_view const& a,
-	std::string_view const& b
-) -> unsigned int
-;
+template<class S, unsigned int stack_buffer_size = 16>
+auto levenshtein_distance(S const& a, S const& b) -> unsigned int {
+	if (a.size() > b.size()) return levenshtein_distance(b, a);
+
+	auto [a_begin, b_begin] = std::mismatch(a.data(), a.data() + a.size(), b.data());
+
+	auto a_end = a.data() + a.size();
+	auto b_end = b.data() + b.size();
+
+	while (a_begin != a_end && b_begin != b_end && *(a_end - 1) == *(b_end - 1))
+	{
+		--a_end;
+		--b_end;
+	}
+
+	if (a_begin == a_end) return b_end - b_begin;
+	auto const buffer_length = a_end - a_begin + 1;
+	unsigned int stack_buffer[stack_buffer_size];
+	auto buffer = buffer_length < stack_buffer_size ? stack_buffer : new unsigned int[buffer_length];
+
+	std::iota(buffer, buffer + buffer_length, 0);
+
+	for (auto i = 0u; i < b_end - b_begin; ++i)
+	{
+		buffer[0] = i;
+		auto temp = i;
+		for (auto j = 0u; j < a_end - a_begin; ++j)
+		{
+			temp = std::min(
+				temp + (a_begin[j] == b_begin[i] ? 0u : 1u),
+				std::min(buffer[j + 1], buffer[j]) + 1u
+			);
+			std::swap(buffer[j + 1], temp);
+		}
+	}
+
+	auto cost = buffer[buffer_length - 1];
+	if (stack_buffer != buffer) delete [] buffer;
+	return cost;
+}
+
+auto levenshtein_distance(char * const a, char * const b) -> unsigned int ;